PaperlayerView exact revision on arXiv

Source-linked AI summary

Hierarchical Edge Computing in SAGSIN: Multi-Layer Network Architecture and Multi-Level Information Processing

Jiajie Xu, Zhengying Lou, Mohamed-Slim Alouini

arXiv:2609.29467v1eess.SY

TL;DR

Raw sensing data are costly to transmit over heterogeneous SAGSIN links, and current approaches do not jointly select processing depth and placement across the full path. The article couples a four-tier architecture with progressive refinement and finds that the energy-minimizing depth shifts with link conditions. Its offshore case study reports longer lifetimes with deeper refinement on costly acoustic links.

  • Problem

    Raw-data delivery across SAGSIN’s heterogeneous links is slow and energy-intensive, while existing approaches do not jointly choose refinement depth and placement across the full path.

    Source: §I · p. 1

  • Method

    The article couples a four-tier SAGSIN architecture with progressive information refinement and characterizes computation, transmission, and storage energy across tiers.

    Source: §I · p. 1

  • Results

    The offshore case study finds that L4 yields the longest lifetime on long acoustic links, while the moderate-cost link’s total-energy minimum is at L3.

    Source: §V · p. 6

  • Takeaways & Limitations

    Refining data can trade modest local processing cost for transmission and storage savings while reducing latency, limiting raw-data exposure, improving delivery reliability, and extending node lifetime.

    Source: §VII · p. 8

    The quantified trade-offs use illustrative parameters, so field validation with real modems and workloads is needed to confirm the predicted gains.

    Source: §VI · p. 8

Abstract

from arXiv · show

Maritime Internet of Things (IoT) deployments increasingly rely on the space-air-ground-sea integrated network (SAGSIN) to connect underwater sensors with terrestrial and space backbones. However, the heterogeneous links along this path, ranging from bandwidth-limited and energy-hungry underwater acoustic channels to high-capacity optical links above the sea, make the transport of massive raw sensing data costly in both latency and energy, and difficult to sustain for unattended, battery-powered nodes. This article presents an edge-computing paradigm for SAGSIN built on two coupled ideas: a Multi-Layer Network Architecture (MLNA) that organizes the underwater, surface, aerial, and ground/space tiers, and Multi-Level Information Processing (MLIP) that progressively refines data from raw measurements toward compact, event-level representations as they ascend the network. We characterize the computation, transmission, and storage energy at each tier and show how distributing refinement across layers trades local processing cost against transmission and storage savings. We then discuss how MLNA-MLIP reduces latency, strengthens data privacy, improves service reliability, and manages energy to prolong network lifetime. A case study on offshore monitoring quantifies the resulting lifetime gains and identifies the optimal processing depth. Open challenges and future directions are outlined.

I. INTRODUCTION

Raw sensing data are costly to carry across SAGSIN’s heterogeneous links, while existing approaches do not jointly select processing depth and placement across the full path. The article couples a four-tier architecture with progressive data refinement.

I. INTRODUCTION

  • Raw-data delivery across heterogeneous links is slow and energy-intensive, threatening the lifetime of unattended, battery-powered underwater sensors.Acoustic links have limited bandwidth and high per-bit energy, while above-sea links also incur costs over long distances.Source: §I · p. 1
  • Existing approaches use fixed node-local processing, leaving unresolved how to choose refinement level and processing tier across the underwater-to-space path.The decision must account for the different energy costs across tiers.Source: §I · p. 1
  • MLIP progressively refines data from raw measurements toward compact event-level information as it travels upward, balancing local processing cost against downstream transmission and storage savings.The framework couples refinement depth with processing placement across SAGSIN tiers.Source: §I · p. 1, §II · p. 2

II. MULTI-LAYER NETWORK ARCHITECTURE

  • MLNA organizes SAGSIN into underwater, surface, aerial, and ground/space tiers with distinct sensing, cross-medium, relay, and aggregation roles.These tiers connect seabed sensing to terrestrial or space-based core networks.Source: §II · pp. 1–2

A. Underwater Layer: Data Collection

The underwater tier collects marine measurements using fixed seabed sensors and mobile AUVs, but its battery constraints and acoustic-link costs make long-term data delivery difficult.

A. Underwater Layer: Data Collection

  • Underwater nodes collect measurements such as temperature, salinity, pressure, and dissolved oxygen; fixed seabed USNs are complemented by AUVs that extend spatial coverage.Fixed nodes are commonly arranged in grid-based topologies for stable, long-term monitoring.Source: A. Underwater Layer: Data Collection · p. 2
  • Fixed USNs are difficult to maintain and typically use 100–500 Wh lithium battery packs that support several months of operation.AUVs can run for tens of hours per charge before returning to the surface for recharging.Source: A. Underwater Layer: Data Collection · p. 2
  • Acoustic links span tens of kilometers but typically provide only 1–10 kbps, with power demand rising with distance and propagation delay arising from sound’s low speed.Multi-kilometer links can require tens of watts; sound speed is 1.5 km/s.Source: A. Underwater Layer: Data Collection · p. 2

B. Surface Layer: Cross-Medium Transformation

Surface buoys bridge underwater acoustics to above-sea links, while aerial HAPS extend coverage and relay data toward ground stations or satellites.

B. Surface Layer: Cross-Medium Transformation

  • Moored buoys provide stable positioning for long-term relay, whereas drifting buoys trade positional stability for flexibility and wider-area coverage.The buoy type depends on the application and environmental conditions.Source: B. Surface Layer: Cross-Medium… · p. 2
  • Surface buoys combine renewable power, usually solar with possible wave or wind harvesting, and onboard batteries for autonomous operation.Hybrid power management and duty cycling can support six months or longer without maintenance.Source: B. Surface Layer: Cross-Medium… · p. 2
  • Nearshore buoys forward data to shore stations, while offshore buoys relay to HAPS; sub-6 GHz links offer above-99% availability and FSO adds Gbps capacity in clear skies.FSO is sensitive to fog, clouds, and sea spray.Source: B. Surface Layer: Cross-Medium… · p. 2

C. Aerial Layer: Connectivity Bridge

  • HAPS bridge surface and ground/space tiers from 17–22 km altitude, remaining quasi-stationary and providing coverage with a radius of at least 100 km.They extend connectivity across wide oceanic areas.Source: C. Aerial Layer: Connectivity Bridge · pp. 2–3
  • HAPS can forward to ground stations or LEO satellites, using sub-6 GHz for robust fallback and higher-capacity mmWave or optical links when conditions and beam control permit.Ground-facing weather and visibility affect link availability and selection.Source: C. Aerial Layer: Connectivity Bridge · p. 3

D. Ground/Space Layer: Data Aggregation

Ground stations and LEO satellites aggregate data and provide backbone access, while hierarchical information levels and energy-aware processing choices determine how much data is refined along the route.

D. Ground/Space Layer: Data Aggregation

  • Ground stations provide stable nearshore connectivity, while mobile LEO satellites offer broad footprints and maintain continuous service through coordinated constellation handovers.Ground-station coverage typically extends 20–40 km; each satellite footprint extends 1,000 km in radius.Source: D. Ground/Space Layer: Data Aggregation · p. 3
  • The four tiers face distinct coordination burdens, from underwater battery and acoustic constraints to surface cross-medium adaptation, aerial link selection, and backbone handovers.The ground/space tier also coordinates resources across domains.Source: D. Ground/Space Layer: Data Aggregation · p. 3

A. Hierarchical Information Levels

  • Information levels begin with raw USN data at L0 and advance through filtering, compression, feature extraction, fusion, or inference, reducing data volume toward application-dependent LN.The total number of processing levels varies by application.Source: A. Hierarchical Information Levels · p. 3
  • Each refinement step adds computation energy but can reduce later transmission and storage costs, so feasible processing depth balances local cost against downstream savings.The framework distributes processing across layers to manage this trade-off and prolong network lifetime.Source: A. Hierarchical Information Levels · p. 3
  • Processing depth can be selected by profiling each refinement step, excluding levels that violate constraints, then choosing the feasible level with the lowest predicted total cost.Runtime inputs include link energy, session overhead, queue state, and residual battery; resource-constrained nodes can use a lookup table.Source: A. Hierarchical Information Levels · p. 3

B. Computation Energy

Progressive refinement reduces data volume but adds computation, so processing depth must account for both the energy cost of computation and the value of avoiding unnecessary work.

B. Computation Energy

  • Each refinement step adds computation energy while reducing data volume, so its depth should balance local processing cost against later transmission and storage savings.The data fractions are representative assumptions; actual ratios depend on data type and processing algorithms.Source: B. Computation Energy · p. 4
  • Computational energy includes arithmetic operations and data movement, with total cost governed mainly by operation count and transferred-data volume.Transfers between main memory and cache can consume substantially more energy than transfers between cache and registers.Source: B. Computation Energy · p. 4
  • Event-triggered processing, sparse computation, lightweight buffering, and in-place processing reduce unnecessary computation or costly main-memory accesses.Event-triggered processing suppresses work when measurements remain stable.Source: B. Computation Energy · p. 4

C. Transmission Energy

Transmission and storage costs depend on both data volume and system conditions; MLIP refinement can reduce these costs, but its benefit must be weighed against local computation.

C. Transmission Energy

  • Transmission energy includes payload delivery and link-establishment overhead such as synchronization, idle listening, handshakes, acknowledgments, and retransmissions.These overheads can be substantial over energy-intensive underwater acoustic links.Source: C. Transmission Energy · p. 4
  • Propagation medium, carrier type, and data volume govern transmission energy, because larger payloads keep the transceiver active longer.Acoustic, radio, and optical links differ in attenuation, data rates, and per-bit energy costs.Source: C. Transmission Energy · p. 4
  • Refining data before transmission reduces payload volume and transceiver active time, but further refinement is beneficial only when transmission savings compensate for added computation.Adaptive power control and link selection can also follow channel and visibility conditions.Source: C. Transmission Energy · p. 4

D. Storage Energy

  • Storage energy can fall when refinement reduces data written and buffered, while application-specific RAM sizing avoids unnecessary retention energy.RAM requires continuous power to retain contents; ROM retains data without continuous power but has higher write-energy cost.Source: D. Storage Energy · p. 4

IV. ROLE OF MLNA-MLIP IN SAGSIN

MLNA–MLIP processes and refines maritime data across network layers, with the stated aims of reducing latency and improving other network functions. Distributed processing and compact representations address queueing and transmission delays, including in offshore seismic monitoring.

IV. ROLE OF MLNA-MLIP IN SAGSIN

  • MLNA–MLIP supports reduced latency, stronger privacy and security, improved service reliability, and energy management intended to prolong network lifetime.Table I summarizes these four stated benefits.Source: §IV · p. 4, Table I · p. 5

A. Latency Reduction

  • Processing at multiple network layers reduces queueing delays from centralized processing, while compact representations reduce transmission latency over bandwidth-constrained links.Distributed processing adds delay, which can be managed by adjusting processing depth to network conditions.Source: A. Latency Reduction · p. 5
  • In offshore seismic monitoring, source or intermediate nodes can extract features and detect anomalies, forwarding event-level indicators instead of raw waveforms.Processed summaries can still reach centralized aggregation and long-term assessment.Source: A. Latency Reduction · p. 5

B. Data Privacy

MLNA–MLIP uses refinement and flexible transmission paths to address privacy, reliability, and energy constraints in maritime IoT. Its examples describe forwarding abstracted data, adapting around impaired links, and balancing computation against transmission energy.

B. Data Privacy

  • Forwarding extracted features rather than sensitive raw data reduces information leakage risk, while shortening wireless transmission time and its temporal attack surface.The abstract representations may lack enough semantic content to reconstruct operational context, without eliminating centralized data aggregation.Source: B. Data Privacy · p. 5

C. Service Reliability

  • Intermediate buffering and refinement, multiple transmission modes, and alternative paths reduce reliance on high-quality links at every hop.In severe weather, buoys can buffer data and switch from unavailable optical links to more robust RF transmission.Source: C. Service Reliability · pp. 5–6

D. Energy Management

  • MLNA–MLIP balances computation and transmission energy along the path; local anomaly detection can suppress redundant reporting and costly acoustic-link wake-ups and handshakes.The offshore oil-and-gas example describes reporting only abnormal summaries to extend node lifetime.Source: D. Energy Management · p. 6

V. CASE STUDY: OFFSHORE MONITORING

The offshore-monitoring case study models how local refinement trades computation against acoustic transmission and storage costs. It finds that lifetime gains and the energy-optimal processing depth depend on link distance and reporting policy.

Case-study setup

  • The representative deployment compares raw L0 real-time transmission with local refinement through L4 and event-triggered reporting, with L1–L3 as intermediate levels.The level-specific data fractions require profiling the sensing modality and processing pipeline; the case-study values illustrate qualitative trade-offs rather than predict absolute lifetimes.Source: §V · p. 6

Simulation setup

  • The simulation sums computation, transmission, and storage energy over a repeating 24-hour workload and defines network lifetime as depletion of 20% of USNs.Transmission includes per-session wake-up, synchronization, and handshake costs plus per-bit acoustic energy; per-bit energy grows with the square of distance.Source: §V · p. 6

The optimal processing depth

  • The energy-optimal depth shifts from L4 on long, costly acoustic links to L3 at moderate distance and shallower levels on inexpensive RF or optical links.Total energy is U-shaped with processing depth: transmission energy falls while computation energy rises, so further refinement eventually costs more than it saves.Source: §V · pp. 6–7

Lifetime versus distance

  • Deeper refinement extends USN lifetime, with the advantage growing as acoustic distance increases.L2 already extends lifetime several-fold; L4 also sharply reduces acoustic sessions, leaving fixed handshake cost as the dominant residual energy use.Source: §V · p. 6, Figure 3 · p. 7

Real-time versus accumulated reporting

  • Event-triggered reporting accumulates and refines data during stable conditions but reports anomalies immediately, balancing routine-session overhead against critical-event timeliness.Batching lowers session count and handshake energy at the cost of higher latency.Source: §V · p. 7

VI. OPEN CHALLENGES AND FUTURE DIRECTIONS

The section identifies deployment-scale challenges for MLNA–MLIP, spanning adaptive depth and cross-layer control to orchestration, representation fidelity, interoperability, security, and field validation.

Adaptive depth selection

  • Total energy is minimized at a depth balancing computation and transmission, but the optimum shifts with link cost, channel state, energy reserves, and traffic.Hand-tuned per-tier rules cannot track these dynamics; lightweight controllers must also fit underwater nodes’ limited compute and energy budgets and prove robust in heterogeneous SAGSIN.Source: Figure 4 · p. 7, §VI · p. 7

Cross-layer link–computation co-scheduling

  • Processing placement and link choice must be co-scheduled across acoustic, RF, and optical links, yet stable distributed schedulers without global state are still missing.The decisions must account for fluctuating link availability, intermittent energy harvesting, and time-varying topology.Source: §VI · p. 7

Complexity-aware orchestration and failure handling

  • Joint optimization across four tiers can impose prohibitive control overhead, while changing channel and energy states can make centralized decisions stale.The proposed direction is hierarchical orchestration: local lightweight refinement and fallback decisions, with coordinators updating cross-tier policies; challenges include bounding signaling, isolating failures, and maintaining service when links become unavailable.Source: §VI · pp. 7–8

Semantic and task-oriented refinement

  • Task-aware representations could reduce transmitted volumes beyond the assumed fractions, but event- and feature-level data must retain sufficient fidelity when raw measurements cannot be recovered.The open questions are how to guarantee downstream analytic fidelity and bound reconstruction error.Source: §VI · p. 8

Standardization, security, and validation

  • Multi-vendor deployment requires interoperable interfaces and lightweight security for refined data, while field validation with real modems and workloads is needed to confirm gains predicted from illustrative parameters.Source: §VI · p. 8

VII. CONCLUSION

The article presents MLNA–MLIP as an edge-computing paradigm for maritime IoT and characterizes its energy trade-offs across SAGSIN tiers. Its offshore-monitoring case study identifies an energy-minimizing processing depth that shifts with the tier and link in use.

Conclusion

  • MLNA–MLIP couples a four-tier SAGSIN architecture with progressively refined information levels for maritime IoT.Source: §VII · p. 8
  • Progressive refinement trades local processing cost for transmission and storage savings while reducing latency, limiting exposure of sensitive raw data, improving delivery reliability, and extending energy-constrained nodes’ lifetime.Source: §VII · p. 8
  • The offshore-monitoring case study identifies an optimal processing depth that minimizes total energy and shifts with the tier and link in use.Source: §VII · p. 8
Loading 2609.29467v1…